README for ASReview-experimental-models: Comprehensive Model Suite

Overview

This plugin for ASReview provides a diverse suite of models, expanding the capabilities of the ASReview software for automated systematic reviews. The plugin integrates several advanced classifiers and feature extraction techniques, enabling users to leverage a broad range of algorithms for their systematic review processes.

! HIGHLY EXPERIMENTAL !

Included Models

XGBoost: A highly efficient and scalable implementation of gradient boosting.

ALL-MPNet-Base-v2 (Hier. Mean): An advanced transformer-based model optimized for semantic understanding with hierarchical mean pooling. Note, max sequence length is 384.

DistilUSE-Base-Multilingual-Cased-v2: A multilingual transformer-based model, offering robust performance across various languages. Note, max sequence length is 128.

LaBSE: Language-agnostic BERT Sentence Embedding model, excellent for semantic similarity and retrieval tasks in multiple languages. Note, max sequence length is 256.

OneHot: A simple baseline model that uses one-hot encoding for feature extraction.

FastText: A powerful text representation and classification model, particularly effective for tasks involving large vocabularies and rich text data. Uses wiki-news-300d-1M-subword.vec.

Convolutional Neural Network: A deep learning model that uses convolutional and pooling layers to extract features from text data.

Doc2Vec (120 vector version): A powerful text representation and classification model, particularly effective for tasks involving large vocabularies and rich text data.

Spacy: Uses en_core_web_md model from Spacy to extract features from text data.

Word2Vec: A powerful text representation and classification model, based on word2vec-google-news-300.

Dynamic_nn: A neural network model that uses a dynamic architecture. Every factor 10 increase in the number of samples, the number of layers is increased.

KNN: A simple baseline model that uses K-Nearest Neighbors for classification.

AdaBoost: A simple baseline model that uses AdaBoost for classification.

Random: A simple baseline model that uses random classification.

Keywords

Classifiers Available:

xgboost
power_cnn
dynamic_nn
adaboost
knn
random

Feature Extractors Available:

ft-sbert
wide_doc2vec
multilingual
labse
fasttext
onehot
spacy
word2vec
scaled-doc2vec
softmax-doc2vec
relu-doc2vec
mxbai

Installation

To install this plugin, use one of the following method:

pip install git+https://github.com/jteijema/ASReview-experimental-models.git

Usage

Once installed, the models from this plugin can be used in ASReview simulations. For example, to use the XGBoost model, run:

asreview simulate example_data_file.csv -m xgboost

Replace xgboost with the appropriate model identifier to use other models.

Compatibility

This plugin is compatible with the latest version of ASReview. Ensure that your ASReview installation is up-to-date to avoid compatibility issues.

License

This ASReview plugin is released under the MIT License. See the LICENSE file for more details.